AI Systems Architect / Lead MLOps

Context

Mentivis designs and operates artificial intelligence systems applied to education, professional training, and skills transformation. We are neither a consulting firm nor a generic integrator: we build sovereign AI infrastructure, in production, for public institutions, enterprise clients, and high-volume training operators.

As part of the rollout of MentivisOS, our AI platform applied to education, we are hiring an AI Systems Architect responsible for designing, industrializing, and operating our multi-LLM pipelines in a sovereign environment.

This is not a prototyping role. The job is to run systems in production, with reliability, traceability, and regulatory compliance commitments to institutional clients.

The role

Orchestration: multi-LLM industrialization

Design and maintain a multi-model orchestration architecture capable of routing, chaining, and falling back between multiple LLMs based on context, cost, latency, and sovereignty constraints. The stack relies on LangChain / LangGraph for agentic chaining, with an abstraction layer enabling model swaps without application refactoring.

Build and operate end-to-end MLOps pipelines: model versioning (MLflow, Weights & Biases), training and fine-tuning orchestration (Ray, DeepSpeed, LoRA/QLoRA for parametric adaptation), model registry with lineage tracking, canary and blue-green deployment on GPU infrastructure (NVIDIA A100/H100, or sovereign cloud equivalents such as OVHcloud, Scaleway, Outscale).

Guardrails, reliability, and governance

Implement a systematic guardrails layer across all LLM outputs: hallucination detection via NLI (Natural Language Inference), content filtering (Guardrails AI, NVIDIA NeMo Guardrails), structural output validation (Pydantic, JSON Schema enforcement), and semantic drift monitoring in production (LangSmith, Langfuse, Arize Phoenix).

Build an automated testing infrastructure for AI systems: regression evaluations on proprietary benchmarks, automated red-teaming, coherence and factual fidelity scoring, with alerting and circuit-breaker mechanisms triggered on degradation.

Ensure regulatory compliance (European AI Act, GDPR, ANSSI frameworks) and produce the technical documentation required for client audits and certifications.

Enterprise integration: LearningOS and TalentOS API

Design and maintain API connectors to LearningOS ecosystems (LMS/LXP: 360Learning, Cornerstone, Moodle, OpenEdX) and TalentOS ecosystems (HRIS/Talent Management: Workday, SAP SuccessFactors, Talentsoft/Cegid) for enterprise deployments.

This covers: bidirectional learner data ingestion (pathways, completions, assessments, certifications), skills framework synchronization (ROME, ESCO, client-specific frameworks), real-time feeding of AI recommendation engines from HR data, and automated generation of contextualized learning content (adaptive learning paths, generated assessments, automatic remediation).

Develop the SDKs and technical documentation enabling client IT departments to integrate our AI services into their existing architectures (REST, GraphQL, webhooks, SSO/SAML).

Infrastructure and observability

Operate the model serving infrastructure (vLLM, TensorRT-LLM, Triton Inference Server) with autoscaling, concurrency management, and throughput optimization (dynamic batching, INT8/FP8 quantization, speculative decoding).

Implement full observability: distributed tracing (OpenTelemetry), inference performance metrics (P50/P95/P99 latency, tokens per second, cost per request), operational dashboards (Grafana, Datadog), and proactive alerting.

Profile

Engineering degree or equivalent, with a minimum of 5 years of production experience on ML/AI systems, including significant exposure to LLM architectures.

Required expertise

Advanced Python (asyncio, strict typing, packaging), TypeScript for API layers. LangChain/LangGraph ecosystem, or equivalent (Semantic Kernel, Haystack). Model fine-tuning and adaptation techniques (LoRA, QLoRA, RLHF, DPO). GPU serving infrastructure and inference optimization. Kubernetes, Docker, CI/CD on ML pipelines. At least one European sovereign cloud provider (OVHcloud, Scaleway, Outscale, 3DS Outscale).

Strong differentiators

Experience on AI Act-compliant or certified projects (ISO 27001, SecNumCloud). Knowledge of the education and training sector and its associated standards (xAPI, LTI, SCORM, CMI5). Integration experience with enterprise HRIS or LMS platforms. Open source contributions in the AI ecosystem.

What we offer

A high-impact position within an operator that deploys AI systems in production for institutional and enterprise clients, not in a research lab disconnected from the field. A lean team where technical decisions are made by those who build. A base in Marseille, at the heart of the Campus Cyber.AI Euromed ecosystem, with direct exposure to European digital sovereignty challenges.

Compensation: based on profile and experience, with a variable component indexed to the results of deployed projects.